AI Models Exhibit Gender-Based Risk Preferences in Financial Decision Making
Large language models significantly alter their financial risk-taking behavior when prompted to assume gender-specific personas, according to a study from Allameh Tabataba'i University. DeepSeek Reasoner and Google's Gemini 2.0 Flash-Lite demonstrated heightened risk aversion when simulating female decision-making—aligning with established behavioral economics patterns.
OpenAI's GPT models maintained neutrality, while Meta's Llama and xAI's Grok produced inconsistent responses. The research employed the Holt-Laury risk assessment framework, revealing how AI systems may inadvertently perpetuate real-world financial stereotypes through their training data.